For a long time, you had to convince voters. Now you sometimes also have to convince an algorithm.
As the 2027 presidential election approaches, we continue to look at an election campaign the way we always have: the candidates, the polls, the rallies, the soundbites, the televised debates, the newspaper front pages, the alliances, the betrayals and the rivalries. All of that still exists. But while we are watching the stage, another campaign is being played out behind the scenes. It fits in our pocket.
Every morning, sometimes before we have even spoken to anyone, we open Instagram, TikTok, Facebook, X or YouTube. We scroll through dozens, sometimes hundreds of pieces of content. And we have the impression that we are choosing what we watch. That is partly an illusion. Because behind our screen, systems have already selected what they think is likely to interest us, make us laugh, worry us, make us angry, or simply stop us from putting our phone down.
The algorithm does not know our convictions the way a friend would. It knows something that is sometimes more useful: our behaviour. It knows which video we paused on for three seconds longer. What we shared. What made us comment. What we watched all the way through. And from these thousands of insignificant little gestures, it learns to show us more of what makes us react.
That is why the 2027 presidential election could be very different from those that came before.
We no longer all watch the same France
Twenty years ago, two French people who were politically opposed could watch the same eight o’clock news and then argue about what they had just seen. They agreed on nothing. But at least they had, for the most part, seen the same images.
Today, two neighbours can live on the same landing, take the same metro, work in the same neighbourhood and vote in the same polling station while being exposed for months to two radically different visions of the country. One will see videos about insecurity every day. Another about purchasing power. A third about immigration. A fourth about the climate. A fifth about Gaza or Ukraine.
Each will naturally end up considering that the subject they see everywhere is the one everyone is talking about. But “everyone” no longer really exists. There are now millions of different news feeds.
This is probably one of the most profound democratic changes of recent years, and we have not yet fully measured its consequences. We have moved from a public space in which we argued around a relatively shared reality to a society in which each person can gradually receive their own version of reality.
The danger is not that the algorithm tells us what to think
It is much more subtle. It can help determine what we are going to think about. And that difference is fundamental.
If I show you images of violence ten times in a day, you will not necessarily adopt the political opinion I expect. But there is a strong chance that security will take up more space in your perception of the country. If your phone shows you, for three weeks, accounts of how hard it is to make it to the end of the month, purchasing power will probably occupy a larger place in your concerns.
This is not automatic manipulation. It is prioritisation. And in politics, ranking a population’s concerns has always been immense power.
For a long time, that power belonged essentially to the media, to political leaders and to institutions. A newsroom decided to put a subject on the front page. A television news bulletin chose to open with one story. A political leader tried to impose their theme on the public debate.
Today, part of this function of prioritisation belongs to automated systems whose workings most citizens almost entirely ignore. And these systems have a particular feature: they live off our attention.
Anger earns clicks
This is perhaps where the problem really begins. An algorithm does not have a political opinion in the human sense of the term. It does not get up in the morning and decide to be on the right or on the left. But it knows how to measure what works.
A nuanced four-minute explanation rarely works as well as a spectacular fifteen-second assertion. Anger works. Fear works. Outrage works. Caricature works. Conflict works extraordinarily well.
And when politics meets this attention economy, the risk is obvious: what is most visible is no longer necessarily what is most important. It is what provokes the most reactions.
A video can be false and still effective. A piece of information can be true but entirely marginal and become omnipresent because it is spectacular. A statement can be taken out of context and become the political subject of the day.
The danger is therefore no longer only the lie. It is also disproportion.
And then artificial intelligence arrived.
From now on, the fake can be industrialised
This is where 2027 could mark a break. Producing a disinformation operation of any real scale once required time, teams and money. You had to write, translate, edit videos, create images and run accounts.
Artificial intelligence radically changes the equation. It makes it possible to create a photograph of an event that never happened. To reproduce a voice. To manufacture a video. To produce hundreds of variations of the same message. To translate a post instantly. To adapt a speech to different audiences. And above all to do all of this at a speed and a cost out of all proportion with what was possible before.
The French authorities are already watching this shift. VIGINUM notes the use of generative AI to manufacture credible texts and images adapted to precise cultural and linguistic contexts, and considers that this technology can structurally increase the information threat. The municipal elections of March 2026 were themselves targeted by four foreign digital interference operations detected and characterised by the French authorities. (SGDSN)
The 2027 presidential election will therefore not be the beginning of this story. It will probably be the first French presidential election to take place once these tools have become accessible to almost everyone.
But the most dangerous fake will probably not be the most spectacular
We all have the famous deepfake in mind: a candidate appears on screen and utters a sentence they never said. That kind of manipulation will certainly be watched. I fear something much more ordinary.
A real video cut ten seconds too soon. A real photograph with a false caption. An authentic statement placed back into a context that is not its own. An exact statistic used to prove something it does not prove at all. An artificial voice added to real images.
Because a crude fake often ends up being identified. A fake mixed with the truth is infinitely harder to fight. And above all, it has a considerable advantage: it takes a few seconds to watch, but sometimes several hours to take apart.
The lie is instantaneous. The truth still has to bring its proof.
And now the revolution we still talk about too little
We are gradually going to ask artificial intelligence to explain politics to us. It is already beginning.
We ask it to plan our trips, explain our contracts, summarise documents, compare products, translate texts or make sense of the news. Why would politics be an exception?
A voter will ask: “Explain this candidate’s platform to me simply.” “What is he proposing for my taxes?” “What did he actually vote for when he was an MP?” “Compare the candidates’ proposals on schools.” And some will inevitably ask: “Which candidate best matches what I think?”
At that moment, something changes profoundly. Artificial intelligence is no longer simply a tool. It becomes an intermediary between the voter and the political debate.
A study published by the Fondation Jean-Jaurès and Onclusive speaks of 2027 as the first genuine French “AI campaign”. It shows in particular that models can rely on sources of very different kinds and reliability, and that they can present as certain information that is wrong, imprecise or out of date.
This raises a question that should concern us collectively: who informs the artificial intelligence that informs us?
Which media does it consult? Which sites? Which sources does it consider credible? Why do some appear more than others? And when false information is reproduced on hundreds of sites, how does the machine distinguish repetition from truth?
After SEO will come the battle to exist in AI answers
This battle is already beginning in business and communications. For twenty years, everyone wanted to appear on the first page of Google. Tomorrow, everyone will want to appear in the answer given by artificial intelligence.
And politics will obviously not escape this logic. Campaign teams will have to understand not only how to speak to journalists, television, search engines and social networks, but also how their platforms, their positions and their candidates are understood and restated by artificial intelligence models.
We could then witness something entirely new: campaigns writing their content not only to be read by humans, but also to be correctly read by machines.
That is why my opening sentence is not, in the end, so provocative: convincing voters will obviously remain the goal. But it may first be necessary to succeed in being visible to the algorithms that accompany them all day long.
And what about us?
That is ultimately the question that interests me most. Because it would be easy to make artificial intelligence the new monster responsible for all our problems. That would be too simple.
An algorithm does not force us to share information without having read it. An AI does not force us to believe an image because it matches exactly what we wanted to believe. A platform does not forbid us from checking elsewhere. We also have our share of responsibility.
Perhaps we will have to relearn something that digital abundance has gradually made us lose: doubt. Doubting an image that is too perfect. Doubting information that immediately makes us angry. Doubting, above all, information that confirms exactly what we already thought.
Because the most effective manipulations are not those that force us to believe something. They are those that tell us precisely what we wanted to hear.
The real battle of 2027 may be the battle for our attention
Among the strategies already observed, VIGINUM describes the artificial amplification of divisive subjects, the feeding of distrust towards the media, and operations intended to alter the perception of a candidate or a party.
It would be excessive to conclude that algorithms will decide the election. The French will vote freely. Their choice will depend on their history, their convictions, their economic situation, their priorities and the political proposals put before them.
But it would be just as naive to treat as neutral the technology that now organises a considerable part of what we see of the world.
The great question of 2027 may therefore not only be: who will the French vote for?
It will also be: what will they have seen before they vote?
Who will have shown them this information? Why will some of it have appeared to them fifty times and the rest never? And how many times will we have confused what “everyone thinks” with what our own phone decided to show us?
The real power of the algorithm may lie there. It will not necessarily tell us who to vote for. It will do something much more discreet: it will accompany part of what we will have seen, read, liked, hated and shared in the months before our choice.
The real invisible candidate of 2027 could therefore very well be the algorithm.
It will have no campaign poster, no office, no rally. It will never debate on a television set. It will appear on no ballot paper.
But it will be in the pocket of almost every voter.
Sihem SouidCEO Edile Consulting